MESSAGE ORGANIZATION AND INFORMATION:
A STUDY IN THE MEASUREMENT OF -
HUMAN INFORMATION PROCESSING ~

Thesis for the Degree of Ph. D.
MICHIGAN STATE UNIVERSI’ T" Y
JOHN SANFORD MICKELSON

1971

III/III!!! III/IIIIII/III/IIII/I/IIII/IIIII/I I

3 1293 10383 4366

This is to certify that the
thesis entitled

MESSAGE ORGANIZATION AND INFORMATION: A STUDY IN THE
MEASUREMENT OF HUMAN INFORMATION PROCESSING

presented by

John S. Mickelson

has been accepted towards fulfillment
of the requirements for

Jh-D- demeinflmunigfljon

 

I 5x ’ 7
$21 217%? A? XXI/7.7K

Mnjor professor

 

0-7639

 

\
I

MESSAGE ORGANIZATION AND INFORMATION: A STUDY
IN THE MEASUREMENT OF HUMAN INFORMATION PROCESSING

By

John Sanford Mickelson

This study was designed to examine relationships between message
organization, message uncertainty, and verbal learning. Message organ-
ization was manipulated by varying the distance between repeated sub-
ject—verb-object triples in a sequence of sentences. Message uncer—
tainty was defined as the uncertainty (variability) of word selections
for subjects responding to messages where certain words were deleted
from the text. Verbal learning was measured by looking at the uncer-
tainty of word selections after subjects had read the intact texts.

The measure, both for uncertainty and verbal learning, was Shannon's
measure for information (H).

Two messages were designed for this study. Message Two was con—
structed with a larger number of lexical choices than Message One.
Message Two was intended to increase the uncertainty of subjects'
responses. Also, for both messages, an introductory paragraph was con-
structed. The introductory paragraph was intended to reduce response
uncertainty.

One hundred and eighty (180) subjects within a 2 x 6 design
responded to one of six variations for each of the two messages.

The results indicated that when messages were presented in a dis-
organized form, uncertainty increased significantly. Organized messages

scored lower on uncertainty whether one looked at total uncertainty.

John Sanford Mickelson

scores or individual sentence scores.

No significant differences were found for messages with and with-
out introductory paragraphs. In some cases, the uncertainty scores
were increased while in others, they decreased.

No significant differences were found between Message One and
Message Two when looking at all message versions. The uncertainty
scores for Message Two were significantly greater than those for
Message One when looking only at organized versions.

The results did indicate a relationship between learning and
processes cf uncertainty reduction. The uncertainty scores obtained

for messages and sentences did provide an index of learning difficulty.

MESSAGE ORGANIZATION AND INFORMATION:
A STUDY IN THE MEASUREMENT OF

HUMAN INFORMATION PROCESSING

By

John Sanford Mickelson

A THESIS

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR.OF PHILOSOPHY

Department of Communication

1971

Ir—\~

Accepted by the faculty of the Department of
Communication, College of Communication Arts, Michigan
State University, in partial fulfillment of the

requirements for the Doctor of Philosophy degree.

 

Director offlThesis

Guidance Committee: Chairman

 

 

 

 

 

ACKNOWLEDGMENTS

To my wife, Stephanie . . . who loved, encouraged and helped
me.

To my Chairman and Director, Dr. Gerald R. Miller . . . who
taught, guided and helped more than he knows.

To my committee: Dr. Erwin Bettinghaus, Dr. Larry Sarbaugh,
Dr. John Gullahorn, and Dr. Eugene Jacobson . . . whose encourage-
ment and criticism was indispensable.

To Dr. James Campbell . . . who was always available when I
needed him.

To Tom Gordon and Ed Wotring . . . because having friends

makes it all possible.

TABLE OF CONTENTS

ACKNOWLEDGMENTS . . . . . . . . . . . . . . .
LIST OF TABLES . . . . . . . . . . . . .
LIST OF FIGURES I I I I I I I I I I I I I I I
Chapter

I RATIONALE AND HYPOTHESES . . . . . . .

II

III

IV

Introduction . . . . . . . . . . .

Information, Uncertainty, and Message Organization .

Information, Uncertainty, and Meaning . . .
Information, Uncertainty, and Language Learning .

METHOD I I I I I I I I I I I I

Operationalizing the Variables . . . . .
Study Design . . . . . . . . . .
Subjects . . . . . . . . . . . .
Procedures . . . . . . . . . . . .
Calculation of Information Scores . . . . .

RESULTS I I I I I I I I I I I I I I
Raw Scores . . . . . . . .
Overall Treatment Effects . . .

Effects of Message Organization (Hypothesis 1) .
Effects of Introductory Paragraphs (Hypothesis 2)
Effects of Semantic Variability (Hypothesis 3)
Relationships Between Uncertainty and

Learning (Hypothesis H) . . . . . . .

DISCUSSION . . . . . . . . . . . . . .

Information Theories and Language Processing .

Message Organization . . . . . . . . .
Paragraph Interrelationships . . . . . .
Semantic Variability . . . . . . . . . .
Uncertainty and Learning . . . . . . . .
Summary . . . . . . . . . . . . . .

iv

iii

vi

vii

16
22
23
23
2A

25

25
25
25
28
28

29

3“

31+
35
37
39
A1
44

Table of Contents (con'd.

BIBLIOGRAPHY . . . .

APPENDIX I . . . . .

APPENDIX II . . . .

APPENDIX III . . . .

Page
H6
50
51

65

LIST OF TABLES

Table Page
1 RAW SCORES . . . . . . . . . . . . . . . . 26
2 FRIEDMAN TWO-WAY ANALYSIS OF VARIANCE

(TREATMENTS x SENTENCES) . . . . . . . . . . 27
3 KRUSKAL-WALLIS ONE-WAY ANALYSIS OF VARIANCE OF DIS-

ORGANIZED MESSAGES COMPARED WITH LEARNING TESTS . . 29
u KENDALL RANK CORRELATIONS: MESSAGE

ONE TREATMENTS x SENTENCES . . . . . . . . . 30
5 KENDALL RANK CORRELATIONS: MESSAGE

Two TREATMENTS x SENTENCES . . . . . . . . . 31
6 KENDALL RANK CORRELATION: MESSAGE ONE

TREATMENTS x MESSAGE TWO TREATMENTS . . . . . . 32

vi

LIST OF FIGURES

Figure Page
I ORGANIZATIONAL SCHEMES: . . . . . . . . . . . l7
(a) MAXIMALLY ORGANIZED
(b) MAXIMALLY DISORGANIZED

2 DELETION SCHEMES . . . . . . . . . . . . . . 21

vii

CHAPTER 1

RATIONALE AND HYPOTHESES

Introduction

 

Many reasons have been given for ngt_applying concepts from
mathematical communication theory to human language behavior. The most
frequent is that the Shannon measure of information does not deal with
meaning. What Shannon (1948) says, however, is that semantic ques-
tions are not relevant to the engineering problems. Note Shannon
does not say that semantic questions are not relevant nor does he say
that information measures are not relevant for semantic problems. As
Rapoport (1953) points out, ". . . because the semantic problems are

not relevant to the technical ones, is not to say the techniques used

 

to solve the technical problems are not useful in solving the semantic
problems." (p. 160) According to Rapoport, then, there is some justi-
fication for exploring how some of Shannon's mathematical concepts might
relate to human language behavior.

Before preceding further, it may be useful to clear up one semantic
problem in the terminology of mathematical communication theory. Mathe-
matical communication theory (Shannon, 1948; Shannon and Weaver, 1957)
views communication as a probabilistic process at least at the engineer—
ing level. At this level, information is defined as the statistical
improbability of a given message with respect to all other messages

which are possible in the language of interest. When considering

application of mathematical communication concepts to human behavior,
it is useful to modify some of the theoretic terms. This need results
from an inconsistency between our own intuitions about what informa—
tion is and does and its mathematical definition as improbability.
What one usually means when labeling a message informative is that
knowledge about a subject or event reduces uncertainty. The use of
the word "uncertainty" in this context equates with the mathematical
communication theorist's use of the word "information."

When it is necessary to refer to the statistical improbability

of an observation or a message, the label uncertainty will be used.

 

When reference is made to what an observer gains from a message, the

label information will be employed.

 

Having clarified how the words information and uncertainty_will

 

 

be used, one can consider more specifically how these concepts can be
applied to human language processes. The present study is specifi-
cally concerned with how information and uncertainty are related to:

(l) the organization of messages, and (2) learning from messages.

Information, Uncertainty, and Message Organization

To understand how people process information in their communi-
cation transactions, it is necessary to adopt a molar view. People
do not communicate phonemes, words, sentences, or grammars in the nor-
mal course of events. Most communication transactions go beyond sim—
ple, single sentence messages. Usually, communication transactions

take the form.of organized sequences of sentences, such as paragraphs

 

or dialogues. The fact of organized sequences implies some guiding

purpose. People organize messages to make them less uncertain for
the receiver, thereby increasing the probability that the correct or
intended meanings will be elicited. This organization is first seen
at the syntactic level.

Studies of relations between syntax and information are limited.
Grammarians argue that the syntax of a language provides an organizing
function for the semantic information in messages. Language syntax has
the effect of constraining or limiting the semantic information by pro-

viding a pre—determined format for presenting semantic relations. This

 

notion of predetermination suggests that the function words of language
might be "overlearned." Thus, if one should delete a syntactic element
from a message, the presented information should increase, but not so
much as if a semantic element were deleted. The basis for this argument
stems from work by Chomsky (1957, 1965) dealing with differences between
the surface and deep structure of language. The surface level of lan-
guage deals with relations between words in sequences. On the deep
structure level (semantic level), words are organized by relational
concepts.

Studies by Miller, Newman and Friedman (1958); Born, Rubenstein
and Sterling (1959); Fillenbaum, Jones and Rapoport (1963); Weinstein,
Feldstein and Jaffe (1965); Tannenbaum, Williams and Clark (1969); and
Hakes and Foss (1970) bear on this point. The results of each of these
studies indicate that for a sample of mutilated text, function words
(articles, auxillary verbs, prepositions, conjunctions) are more
easily replaced than are semantic words (nouns, verbs, adjectives,
adverbs). Most of these studies are based simply on the accuracy of

of an observer's responses to a mutilated text. The Tannenbaum,

gt_313 study is more interesting. The researchers were interested in
determining whether providing an observer with a form class designator
for each deletion would improve his performance. The results were
affirmative. For deletions from semantic form classes, observers more
frequently chose words from that form class. Verbatim replacements in
the semantic.form class remained the same. With function words,
observers performed better both at choosing a word from the correct
form class and also with making verbatim replacements.

The study by Hakes and Foss was constructed to test the utility
of relative pronouns (function words) as cues to the underlying deep
structure for self embedded sentences. Observers heard sentences,
half of which had the relative pronouns deleted; e.g., "The car the
man the dog bit drove crashed." These were mixed with sentences which
retained the relative pronoun; e.g., "The car that the man whom the
dog bit drove crashed." According to two measures of sentence compre-
hension — phoneme monitoring and sentence paraphrasing - observers pro-
cessed sentences retaining the relative pronoun more effectively.

These and other studies (Miller and Friedman, 1957; Ruhanstein
and Pollack, 1963; Pollack, 1963; Gross, 1966; Butler and Merikle, 1970;
Postman, 1970; Swanson and Wickens, 1970) indicate that language syntax
performs an organizing function on language which serves to reduce the
presented uncertainty of messages. It is not unreasonable to suppose
that the sequencing of sentences in paragraph form serves a similar
function.

Neither grammarians nor communication scientists have considered
at length the interrelationships between sentences. Consider the par-

agraph sequence you are now reading. Sentence 1 provides a set for the

reader to think about what will follow. Under one form of organiza-
tion, each succeeding sentence should operate to reduce the reader's
uncertainty about what is meant by that first sentence. To the extent
that the organization of this paragraph is effective, the reader's
uncertainty about what is meant by the relations between sentences in
sequence and how they operate to reduce uncertainty should be removed
by the end of the concluding sentence.

Although the question of sentence sequencing for paragraphs has
not been studied in any depth, one can speculate about the effect on
an observer of destroying paragraph structure. Suppose one takes a
sequence of paragraphs and reprints them without indenting. If the
original paragraphs (sentence sequences) had the effect of reducing
uncertainty, it is reasonable to expect that observers could locate
and recognize paragraph boundaries. A study by Koen, Becker and Young
(1969) obtained results in line with this prediction. Two experiments
were conducted to determine: (1) whether observers could recognize
paragraph boundaries in unindented prose, and (2) whether the cues to
paragraph boundaries were formal (grammatical) or semantic. For the
second study, prose passages were replaced by nonsense paralogs.
Observers were able to recognize paragraph boundaries in both prose
and nonsense versions with reliabilities .86 and .75 respectively.

Another study by Darnell (1963) examined the relation between
sentence order and comprehension. A message of fifteen sentences was
constructed so as to have a deductive order. The original order of
these sentences was manipulated to obtain six messages each of which
presented an increasing degree of disorder. Certain items were deleted

from each message and 83 were asked to replace the deleted items. The

mean cloze scores for the seven message forms were found to be signi-
ficantly different. Also the degree of disorganization was positively
correlated to the difficulty of replacing deleted items.

Both of these experiments attribute importance to the organized
structure of sentences. The Darnell study indicates this organization
might have something to do with the statistical pr0perties of messages.
Clearly, the organization of messages does affect the capacity of peo-
ple to deal with those messages. The less uncertain the organization
of a message, the more probable it is that a receiver will gain the
information intended by the producer of the message.

The arguments presented in this section give rise to the first
hypothesis for this study:

H1: The uncertainty presented by an organized

sequence of sentences will be significantly
less than the uncertainty of the same sentences
in a disorganized sequence.

The first hypothesis raises questions only about the organization
of sentences within a paragraph. One can also ask what is the relation
of one paragraph to another. Hypothesis 2 is:

H2: There will be a significant decrease in the

uncertainty of a message when it is preceded
by a message intended to reduce its uncertainty.

Information, Uncertainty and Meaning_

 

To deal with problems of semantic information, it is necessary
to think of what a human source does - functionally - when he sends a

message. Certainty the source's primary goal is not to have the receiver

accurately reproduce the words of the message. More probably the source
selects and transmits words that have some probability of eliciting s e-
gifig_meanings from the receiver. Thus one would argue after MacKay
(1969) that measurement of semantic information is a selective function
defined not on words but on meanings. The measurement of semantic
information should be based not on the relative frequency of words, but
on the relative frequencies of word meanings.

The difficulties of assessing information at the semantic level
are compounded by problems unsolved at other levels of language. It
becomes necessary to deal not only with the informational structure of
the language itself but also with the whole semantic space of persons
using language to communicate. One must deal with both the denotative
and the connotative aspects of meanings of words. One must deal not
only with contextual relations between words, but also with the context
within which the words are spoken or written. At the semantic level,
the statistical structure of language is less dependent upon how it is
represented in random samples of literature than upon an individual's
meanings for words alone or in sequence.

Studies of language acquisition illustrate the conceptual rele—
vance of information theory at semantic levels. Luria (1969, 1970)
reports research showing that when a child begins to use a word, he
applies the word to anything having a central attribute of the word's
referent. The child might apply the word "car" to any transportation
mode. At first the central feature of car is that it moves or carries.
Thus the word "car" might apply to horses, trucks, trains or airplanes.
As the child acquires knowledge of "car," he becomes more discriminat-

ing in his use of the word. The word begins to regulate his behavior.

The child begins to use the word ”car" to discriminate cars from horses,
trains and airplanes even though he may yet apply the word to trucks.
But, in general, it seems evident that with knowledge acquisition, the
variability (uncertainty) between word use and referent is decreased.

While there is reason to suspect that the uncertainty presented
by messages is a function of semantic variability, few attempts have
been made to assess the amount of semantic information in messages.
Rosenberg (1969) approaches the problem of semantic information in sen-
tences by distinguishing between semantically well integrated (SWI) and
semantically poorly integrated (SPI) sentences. He argues that where
sentences are semantically well integrated, observers will be better
able to recall the accompanying verbal material. The two types of sen—
tences are distinguishable by sentence pairs like: (1) The doctor who
fired the janitor cured the patient (SWI), and (2) The doctor who fired
the janitor shoots the author (SPI). Again, no attempt was made to
construct a measure of semantic integration. Nevertheless, on princi—
ple, one possible measure is obvious. One can simply count the number
of poor integrations and compare this with the number of good integra—
tions. The chief problem or disadvantage of this measure is that it is
based on a definition of semantic integration that would be unlikely to
apply in normal use of language. The SPI sentence above simply has a
low probability of occurrence.

A more satisfying approach to the semantic complexity of sen—
tences is taken by Perfetti (1969). Perfetti employed two measures of
lexical density for sentences. One measure was based on the absolute
number of semantic words as compared with the number of function words

for sentences. The second measure was based upon the number of items

of Specific information offered by a sentence. The sentence "The fire-
men hurried to the factory that was reported burning" contains three
items of information. They are "firemen hurried to factory," "factory
was burning," and "someone reported it." This sentence would be con-
sidered to have a lower lexical density than the sentence "The museum
is providing free public lectures on modern art." Five separate items
of information are presented by this statement.

The study by Perfetti was designed to determine whether the se-
mantic or the syntactic complexity of sentences best predicts diffi-
culties people have learning sentences. His findings indicate that
increasing either syntactic or semantic complexity increases learning
difficulties. The measures of semantic difficulties, however, provided
better predictions of learning difficulty.

If Luria's report is taken in conjunction with Perfetti's research,
there is some basis for arguing that the critical aspect for the sta-
tistical structure of language is related to the meanings people have
for words. Thus one would expect that the organization of messages
Operates somehow on the semantic structure to increase the probability
that specific meanings are selected by the receiver. Further, one
would expect that to the extent message organization does not increase
the probability for specific meanings, learning difficulties will be
increased.

The preceding arguments give rise to the third hypothesis for
this study.

H3: If the lexical variability of one message is greater

than that of another, uncertainty will also be
significantly greater.

10
Information, Uncertainty and Language Learning

To understand how people learn from messages, it is necessary to
consider the process by which information is gained from messages.
Behavioral scientists (Landahl, 1941; Ammons, 1956; Rapoport, 1956;
Lenneberg, 1957; Kochen and Galanter, 1958; RapoPort, 1961; Attneave,
1967; Moss and Neidt, 1969; Malic, 1970) have argued that learning can
be viewed as a process of uncertainty (entropy) reduction. Ammons
suggests that learning outcomes be thought of as uncertainty reduction
since information is what removes uncertainty. Malic contrasts physical
and intellectual systems. The physical environment constrains Open sys-
tems such that they tend to reach a stationary state with a minimum of
entropy production. Intellectual systems, in contrast, impose constraints
on the environment. Intellectual systems consume entropy (uncertainty).
By consuming entropy the intellectual being acts against the environ-
ment, changing relations in the environment and heightening levels of
organization. These views imply that learning is a process of consum-
ing or removing presented uncertainty. In general, research supports
this view.

A series of studies by Landahl (1941a, 1941b) and Rapoport (1956,
1960, 1961) serve to illustrate the efficacy of learning defined as
uncertainty reduction. The Landahl studies (though pre-mathematical
communication theory) were designed to test a theory of error elimina—
tion based on known properties of neural mechanisms. Landahl predicted
both error rate and number of errors made in a learning task would depend
upon the number of elements to be learned. One hundred and twenty pairs

of four letter words were prepared in groups of A, 8 and 12. Subjects

11

were told that for each stimulus word, there was one response word with
which the stimulus word should be associated and that they were to
determine the correct word by trial and error. This procedure was
carried out to 16 trials, of which the last three were errorless except
in two cases. The four word pairs were perfectly learned after six
trials, eight word pairs after 10 and 12 word pairs after 14. Both
error rate and number of errors increased with number of items to be
leaned.

Landahl's findings led Rapoport to inquire whether learning was
a simple function of number of elements or of the statistical structure
of the language; and, if so, could the relation be expressed in infor-
mation theoretic terms? Rapoport (1960) taught several groups of sub-
jects four artificial languages. Subjects associated irregular geo—
metric figures with their respective names in the language. The refer-
ents were the same in all four language, five different geometric
figures and five different colors. Of 25 possible combinations, 20
were used as referents. Referents were presented in random order and
subjects required to make a correct verbal response depending on the

language being learned. Languages were structured as follows (p. 88):

L1: With each of the 20 figures a single syllable
was associated.
L2:, With each of the 20 figures a pair of distinct

syllables was associated, no syllable being
associated with more than one figure.

L : With each of the five shapes and five colors
a single syllable was associated. The shape
always came first.

L : The vocabulary of this language is the same as L
except there is no systematic correspondence between
syllables and aspects (color and shape) of the referents.

12

The differential structure of these languages gives rise to the follow-

ing uncertainty calculations:

H(Ll) = u2.3 nits
H(L2) = 84.6 nits
H(L3) = 10.3 nits
H<L4) = 65.1 nits

Rapoport argues that the rate of increase of information should be a
function of learner errors. Thus if H is the total uncertainty of the
task, the information gained (h) should be equal to initial uncertainty
(H) less remaining uncertainty (U) where U is the probability of a cor—
rect response to the stimulus. Data obtained by Rapoport confirm his
prediction.

A more recent study by Hyman and Kaufman (1966) provides further
confirmation for the Landahl and Rapoport findings. Two languages were
constructed. A low infOrmation language used eight black geometric
forms as symbols. The high information language was constructed by com-
bining four colors with each of the eight figures. Under this construc-
tion, the low H language yields 3 bits per symbol and the high one 5
bits per symbol. Messages of four, six and eight symbols were constructed
for recall by subjects. The number of symbols recalled by subjects was
shown to be constant under all conditions and a function only_of bits
per symbol. Message length was found to have no effect on recall.

These studies of language behavior indicate that learning is a
function of the statistical characteristics of the language. The lan-
guages used differ in several respects from natural language. The lan—

guages are artificial. The language words and their respective refer-

13

ents are artificial. Words and referents have a one-to-one correspon-
dence. It is assumed that the subjects have no prior knowledge about
the language. Subjects learn from ground zero, so to speak. And the
languages are simple enough that they can be ”almost" perfectly learned.
With natural language, few of these assumptions are met. The words and
referents of natural languages do not have a one—to—one correspondence.
Only rarely is one able to deal with people having no §_priori knowledge
of their language. In other words, these studies imply that as people
learn language, language uncertainty is removed completely. Certainly
this seems an untenable conclusion. In fact it can be argued that it
is the indeterminancy of languages which makes them work so well.
Suppose one thinks of words as having some specified frequency dis—
tribution over specific stimulus continuums; in other words, the lan—
guage retains some degree of indeterminancy or uncertainty. What is
true, say, of the frequency distribution of color words over the visi—
ble light spectrum? Lenneberg (1957) designed a study to answer this
question. Subjects were given the mental set for a language learning
task. Three groups of subjects were to learn a foreign language of four
words, each of which referred to a color. One group learned a language
where the degree of uncertainty expressed by the color words was simi-
lar to English. Two other groups learned languages with a degree of
uncertainty less or greater than that of English. The language expres-
sing less uncertainty was, at first, more difficult than English but
than learning proceded at a more rapid rate. The high uncertainty lan-
guage was most difficult to learn and the rate slowest. Thus the Lenne-
berg study emphasizes the importance for learning of a language's statis—

tical structure while accounting for two additional factors: (1) the

l4

indeterminant nature of language, and (2) the user's knowledge of his
language.

These three studies highlight the significant, perhaps critical,
role played by the statistical nature of a language in learning. The
results suggest that peeple learn the statistical properties of lan—
guage and that people are conditioned by their state of knowledge to
expect certain degrees of indeterminancy. Any attempt, then, to assess
people's information processing behavior must measure both initial and
final uncertainty. As Brillouin (1957) points out, the informational
impact of a message can be determined only if we know the field of
uncertainty before and after an observation. If final uncertainty is
small, the obtained information is large.

The learning of languages seems to be closely related to their
underlying statistical structure. This raises the question expressed
by the fourth hypothesis for this study:

H4: There will be a significant positive correlation

between the uncertainty presented by items of

information in a message and the difficulty of
learning these items.

Summary

The rationale described above raises several questions relevant
to message transmission in behavioral settings. One question has to do
with relations between message organization and message uncertainty.

It is argued that message organization is purposeful and has the func—
tion of reducing message uncertainty. A second question has to do with

to what extent sentences transmit information. If a sentence takes its

 

if [I‘ll

15

"natural" place in an organized sequence, will it serve to reduce the
uncertainty of the sentences or paragraphs which follow it? Thirdly,

to what extent is the uncertainty of a message affected by altering

the semantic variability? And, finally, can it be shown that the uncer-
tainty structure of messages is related to problems of learning

message contents?

CHAPTER 2

METHOD

The present study deals with relationships among three varia—
bles: message organization, message uncertainty, and learning dif-

ficulty.

Operationalizing the Variables

Message Organization: Perfetti's concept of a semantic partial
as the "smallest" unit of semantic information for a sentence can serve
as a basis of at least one kind of paragraph structure. The semantic
partial is essentially a subject-verb-object (S-V-O) triple where any
of the S-V-O units may be a phrase rather than a single lexical item.
Any sentence can be composed of an indefinite number of semantic par-
tials which can be linked one to another in a variety of ways. One
such linkage would be to exchange the O of partial A with the S of par-
tial B yielding a chaining effect like A = task concerns concept and
B = concept is uncertainty. That is, the atomic units of each seman-
tic partial can be made the linkage unit within sentences. In some-
what the same manner, one might make the semantic partials the linkage
units between sentences.

Two messages were designed for this study. Each message con-
sists of ten sentences. Each sentence contains four semantic par—
tials which are represented by the capital letters in Figure la.

The organized paragraphs used for this study follow the pattern

16

 

III. ii I.‘{[

17

 

 

 

 

 

(a) (b)
ABCD EFGH
BCDE GHIJ
CDEF CDEF
DEFG IJKL
EFGH ABCD
FGHI JKLM
GHIJ BCDE
HIJK HIJK
IJKL DEFG
JKLM FGHI
Figure 1: Organizational Schemes:
(a) Maximally Organized
(b) Maximally Disopganized
depicted in Figure la. Each semantic partial (A, B, . . . , M) repre-

sents an assertion which is to be communicated by the sentences of the
paragraphs. The individual assertions are then further related one to
another by embedding the partials in a normal sentence grammar.

Maximum organization is achieved in this case by writing the
paragraph sentences such that each sentence repeats three partials
introduced by the previous sentence. Although this represents an
extreme form of organization, it is a form where the relationships
among sentences are easily seen and controlled.

The above plan imposes a certain level of redundancy on both sen—
tence and paragraph structure. In particular the structure of a par-
agraph can be accurately specified and manipulated. Manipulation is
achieved by varying the nature of linkages at either the paragraph or
the sentence level. One can, for instance, define maximal disorgani-
zation for the above paragraph pattern by unfolding the pattern, by
maximizing the distances between related sentences.

The pattern shown

in Figure 1b achieves this purpose.

18

The two messages designed for this study, along with the required

introductory paragraphs, are in both an organized and a disorganized form.

Message One: Introductopy Paragraph

The study of human communication draws upon concepts from a variety of
different disciplines. Among these are concepts like organization,
memory, information, learning and so on. The task in which you are
here participating is concerned with generating data to help assess
the relationships between several of these concepts and uncertainty.

Deletion* Paragraph: Opganized

1This task involves the conceptA uncertainty and its reduction
by information. 2The uncertainty conceptB states that uncertainty
will be reduced as information is gained. 3We assume that uncertainty
will beqreduced as information is gained according to some mental pro-
cess. Generally, the reductéon takes place as information is gained
in the process of learning. Infogmation is gained in the learning

process as one learns patterns. That which is gqined in the process
of learning pattern is the organization of stimuli . In the process
The

of learningG, patterns organize stimuli into code sequences.

learning of patterns involves the organization of stimuli into code
sequences like sentencesK. gThis patterning or or anizing of stimuli
into code sequences:1 like sentences facilitates transmission of infor-
mation. 10Thus, the organization of stimuli i to code sequences facil-
itate information transmission and uncertainty reduction.

 

 

 

 

 

 

 

 

Deletion Paragpaph: Disorganized

 

6That which is gaiped En the process of learning pattern is the
organization of stimuli . Generally, the reduction takes place as
information is gained in the process of learning. 8The learning of
patterns involves the organization of stimuli into code seqpences like
sentences . The uncertainty concept states that uncertainty will
be reduced as information is gained. 10Thus, the organization of
stimuli intfi code sequencis facilitates information transmission and
uncertainty reduction. This task involves the conceptA uncertainty
and its reduction by information. This patterning or or anizin of
stimuli into codensequencesJ like sentences facilitates transmission
of information. 0We assume that uncertainty will be reduced as infor-
mation is gainedB according to some mental process. In the process
of learning , patterns organize stimuli into code sequences. Infor-
mation is gained in the learning process as one learns patterns.

 

 

 

 

 

 

 

 

NNumeric superscripts refer to organized sentence order; alphabetic
superscripts refer to the semantic partial deleted from the sentence.

19

Message Two: Introductory Paragraph

The task in which you are here participating is concerned with gener—
ating data to help assess the relationships between several communi—
cation concepts and uncertainty. Among these are concepts like "code,"
information, learning and so on. The study of communication draws

upon many of these concepts.

Deletion Paragraph: Organized

 

lCommunication involves the study? of how pgople use symbols or
codes organized into sequences like sentences. Peo 1e use symbolsB,
codes and organized sequences or sentences in information transactions.
3When codes are organized in sequences like sentences for ipformation
transactions , the resulting messages produce uncertainty. “The
sequences or sentences used in information transactions produce uncer—
tainty or semantic ambiguity. 5Thus, information transactions invole
messages which produce uncertainty and create semantic ambiguity which
message receivers must resolve. 6Message uncertainty or semantic
ambiguity must be resolved by the messa e receiver before understand-
ipg_can occurI. 7A3 semantic ambiguity is resolved by tge message
receiver, understanding occurs and meaning is acquired. Through
message receiver resolution, understanding occurs, meaning is acquired
and message contents are learnedK. With the occurrence of under—
standing and meaning acquigitionJ by learning of message contents,
uncertainty is reduced. tUAs meaning is acquired through learning of
message contents and reduction of uncertainty, communication is achieved .

 

 

 

 

 

 

 

 

 

Deletion Paraggaph: Disorganized

 

6Message uncertainty or semantic ambiguity must be resolved by
the message receiver before understanding can occurI. uThe sequences
or sentencesD used in information transactions produce uncertainty or
semantic ambiguity. 8Through message receiver resolution, understanding
occurs, meaning is acquired and message contents are learnedK. 2People
use symbols , codes or organized sequences, or sentences in information

 

transactions. As meaning is acquired through learning of messa e con-
tents and reduction of uncertainty, communication is achieved . Commun-

 

ication involves the study? of how people use symbols or codes organized
into sequences like sentences. 9With the occurrence of understanding

and meaningacquisitionJ by learning of message contents, uncertainty

is reduced. 5When codes are organized in sequences like sentences for
information transactiops , the resulting messages produce uncertainty.

7A3 semantic ambiguity“ is resolved by the message receiver, under—
standing occurs, meaning is acquired and message contents are learned.
5Thus, information transactions involve messages which produce uncer—
tainty and create semantic ambiguity which message receivers must resolve.

 

 

 

 

The semantic partials (S-V-O) triples from which these messages were

20

derived are found in Appendix I. The messages differ only in terms of
lexical selections for the partials and the linkages among the partials.
Message One is designed to present less uncertainty than Message Two.
Therefore, semantic partials of Message One are linked by using the
word taking the object position in partial A as the word for the sub-
ject position for partial B. For Message Two, lexical choices are
independent.

Message Uncertainty: Numerous attempts have been made to assess
the uncertainty of messages or observations. In general, these mea—
sures are based on some modification of Wilson Taylor's (1953, 1954)
plpgp_procedure. Studies by Born, Rubenstein and Sterling (1959);
Fillenbaum, Jones and Rapoport (1963); Treisman (1965); Gross (1966);
Darnell (1968); Mickelson (1968); Tannenbaum, Williams and Clark (1969);
Hakes and Foss (1970) and Butler and Merikle (1970) have used this
technique. Ss are presented with some portion of text from which words
have been either systematically or randomly deleted. The distribution
of words for a deletion space is converted to a probability distribu-

tion and measured according to the formula:
U = -f p(i) log p(i)

where p(i) is the probability of occurrence of a particular word in a
deletion space.

One advantage of this method is its flexibility. Textual dele—
tions can be manipulated according to the precise interests of the
investigator. If the investigator is only interested in informational
properties of certain form classes, then only words from these classes

need be deleted. If the investigator is only interested in the infor-

21

mational properties of a specific semantic dimension, he need delete
only those lexical items pertaining to that dimension.

A deletion scheme which systematically deletes words according
to how sentences within paragraphs are related was constructed for
this study. Because the central interest of this study was with
words of semantic form classes, only words falling into those classes
were deleted. Also because it was necessary to be able to compare
information values for sentences, the same number of words were deleted
from each sentence. Accordingly, the deletion scheme for this study
was based upon the organization of semantic partials. From each sen-

tence, one semantic partial was deleted according to the pattern

 

Organized Deletion
Pattern Selection Distance
1. ABCD A P
2. BCDE B l
3. 'CDEE E I
u. DEFG D 3
5. _EFGH F 2
6. FCHI I 0
7. GHIJ G 3
8. —HIJK k 0
9. IJRL J 2
IO. JKLM_ M 0
Figure 2: Deletion Schemes

 

shown in Figure 2. The same deletion scheme was used for both mes-
sages. The column labeled Distance in Figure 2 refers to how many
times a partial is mentioned in the text before it is deleted.
Measure of Learning Difficulty: A test was designed for each
message in order to assess the difficulty people have learning the

contents of each message in intact form. To make this learning data

22

comparable with the uncertainty data in the mutilated or deleted ver-
sions of the messages, a completion type test was constructed. Each
question was constructed to call for the information represented by

one semantic partial. One question for Message One is:

6. The paragraph describes a involving

 

a certain .

 

Test questions for this study have very nearly the same form as the
mutilated message forms. This similarity is necessary to make test
scores comparable with uncertainty scores. The complete test for

each message is found in Appendix II.

Study Design

 

Six treatments were designed for each of the two messages, mak—
ing a 2 x 6 design. In two treatments, Ss were presented with mutilated
versions of the test messages, one treatment group receiving the organ-
ized version and another the disorganized version. In the second two
treatments, the organized and disorganized versions of the test mes-
sages were preceded by an introductory paragraph. The last two treat-
ments received the intact messages and test questions — one group
receiving the test message only and the other the test message pre—
ceded by an introductory paragraph.

Given this design, there are twelve different message forms,

six for each message. These message forms are found in Appendix III.

23

Subjects

83 were 180 graduate and undergraduate students enrolled either
in courses in the Department of Communication or report writing classes
in the Department of Office Administration and Law at Michigan State

University. 83 were randomly assigned to treatment groups.

Procedures

 

88 were given the test instruments in a normal classroom
setting. Ss were told that the experiment was a study of language
behavior and asked to read the instructions carefully. Those 83 who
received the mutilated messages were instructed as follows:

This booklet contains a paragraph from which several

words have been systematically deleted. Your task is

to try and replace the missing words.

You are not expected to fill in every blank with the

exact word that has been left out. Your task is to

fill in each blank with a word which seems to fit in
the context of the paragraph.

 

 

Do your best to fill in all blanks. If nothing seems
to fit in a particular blank, then guess.

 

 

When a mutilated message was preceded by an introductory paragraph, 88
were asked to read that paragraph before filling in the blanks. 88

receiving the intact messages were instructed as follows:

On page 2 on this booklet you will find a one—paragraph
message. Read this message through carefully and then
turn to page 3.

On page 3 you will find a set of 10 completion-type
questions which can be answered on the basis of the
paragraph you have just read. Without turning back
to that paragraph, try and fill in the missing wordso

24

You are not expected to fill in every blank with
the exact word that has.been left out, but do your
best to fill in each blank. If you can recall
nothing that seems to fit a particular blank, then

guess .

 

 

On the basis of a pre-test, it was found that most people were
able to complete the task in ten to fifteen minutes. As the end of
fifteen minutes, the experimenter asked the SS to finish quickly and
stressed filling in all blanks. After another five minutes, the

instruments were collected.

Calculation of Information Scores

 

Instruments were scored by writing Ss responses for each dele—
tion space on an IBM code sheet. A frequency distribution for the
responses to each blank was then obtained. The frequency for each
word was then divided by 15 (15 is the n for each treatment group) to
obtain the probability of a particular word in a blank. These proba—
bilities were converted to information scores for each word by con—
sulting a table of -p(i) log2 p(i) (Staniland, 1966, pp. 207-209).
The information scores for each word were added and used as the score
for a particular blank. The scores for each blank in a sentence were
added and used as the score for that sentence. Sentence scores were

added to obtain a score for the message under a particular treatment.

CHAPTER 3

RESULTS

Raw Scores

 

Raw scores obtained according to the study procedures are
found in Table 1. These are uncertainty or information scores, one
for each sentence (Sl . . . 810) of each message (Message One and
Message Two) under each treatment (Tl . . . T6; T7 . . . T12). These
scores were subjected to several nonparametric analyses, the results

of which are described here.

Overall Treatment Effects

 

Before preceding with analyses of the specific hypotheses for
this study, scores obtained under the different treatments were ana-
lyzed using a Friedman two—way analysis of variance (Table 2). The
Friedman test makes it possible to determine whether sentence scores
obtained under the six treatments are drawn from the same or from dif—
ferent populations. Table 2 shows that the treatments for each message
did have differential effects as predicted. Treatments were signifi-
cantly different for both messages. (The parametric equivalent of

the Friedman analysis is found in Appendix III).

Effects of Message Organization (Hypothesis 1)

 

To test the effects of message organization, the total uncertainty

25

26

 

 

 

 

 

 

 

 

 

 

 

TABLE 1: RAW SCORES
TREATMENT O D OI DI L‘ LI
T1 T2 T3 T4 T5 T6
81 6.1560 6.8046 4.5059 5.1536 4.1730 5.0712
Sentences: S2 6.0725 5.6584 4.9762 5.3781 2.6253 4.1787
Message One 83 2.3816 4.9631 3.4672 3.2467 2.4140 2.9572
(Low S4 1.7343 5.2233 4.2323 4.7227 1.0630 0.7212
Uncertainty) 85 1.7412 4.1002 2.6874 4.7324 2.5130 2.5894
86 6.1913 10.6986 8.1268 10.0452 3.5121 5.9056
S7 3.7270 5.3669 4.2392 5.1804 4.0013 4.4180
S8 6.1737 10.5584 5.3426 8.3459 3.3459 5.5017
89 8.1663 12.4028 8.3164 9.0890 5.7498 5.4057
810 0.7055 5.2330 2.0892 4.1903 0.3575 1.0630
Ml (Total un—
certainty) 43.0994 71.0093 47.9837 60.0843 29.8659 37.8117
T7 T8 T9 T10 T11 T12
81 4.3710 4.9407 3.4133 4.6440 4.5270 5.0976
Sentences: S2 2.6092 5.3461 3.9934 6.5394 3.2121 4.4326
Message Two S3 5.5234 8.5867 7.0769 6.8833 4.3794 3.5212
(High S4 4.6511 6.6505 4.0666 5.5837 5.9220 4.5504
Uncertainty) SS 4.4650 6.1994 5.3182 4.1360 4.0167 2.0032
86 4.7318 6.2334 4.4591 6.1556 4.9103 4.6079
87 2.9501 7.1637 4.3712 6.1127 3.5924 3.3009
88 8.2291 6.5602 9.9665 10.3256 5.1758 6.4215
89 5.5629 4.0034 5.6727 7.0698 6.4164 6.0434
810 6.8370 4.8112 5.1927 5.9704 5.2362 5.4191
Ms (Total un- " ‘
certainty) 49.9306 60.4904 53.5306 63.4205 47.3883 45.4978
Key:
0 = Organized OI = Organized with introductory paragraph
D = Disorganized DI = Disorganized with introductory paragraph
L = Learning Test
LI = Learning test with introductory paragraph

27

scores for organized (T1, T3, T7, T9) and disorganized (T2, T4, T8,
T10) versions were combined in a single distribution and ranked. It

is clear from Table 1 that the uncertainty scores obtained when mes-
sages were in organized form were in all cases lower than scores
obtained from disorganized messages. A Mann-Whitney test of the asser-
tion that organization of sentences reduced uncertainty was signifi-

cant (U = O, P < .05).

TABLE 2: FRIEDMAN TWO-WAY ANALYSIS OF VARIANCE
(TREATMENTS x SENTENCES)

 

 

 

 

Message One Message Two
(rank sums) (rank sums)
Organized (0) T1 29 T7 29
Disorganized (D) T2 59 T8 45
Organized with
introduction (OI) T3 34 T9 30
Disorganized with
introduction (DI) T4 47 T10 48
Learning test (L) T5 15 T11 29
Learning test with
introduction (LI) T6 26 T12 29
Xr2 = 49.444 25.55*
*p, < . 001

Stronger support for the hypothesis that organization reduces
uncertainty can be shown by demonstrating that for each organized/dis-
organized pair of sentences, the disorganized sentence results in a
larger uncertainty score. Such a demonstration can be achieved by
applying the sign test to the 40 organized/disorganized sentence pairs.

Of these 40 pairs, in only 7 cases does the organized sentence score

28

exceed its disorganized counterpart (x = 7, p <:.001).

These two analyses provide strong confirmation for the hypothesis
that the uncertainty presented by an organized sequence of sentences
will be significantly less than the uncertainty of the same sentences
in a disorganized sequence. Not only are the uncertainty totals for
each organized message lower than for disorganized messages, but also

the individual sentence scores are significantly lower.

Effects of Introductory Paragraphs (Hypothesis 2)

 

The second hypothesis of this study has to do with the effects
of introductory paragraphs. Specifically, it has been argued that
there will be a significant decrease in the uncertainty of a message
when it is preceded by a message intended to reduce its uncertainty.
Uncertainty scores obtained for message versions with and without intro-
ductory paragraphs were compared. A Mann-Whitney test applied to
message totals obtained for these message versions was non-significant.
A comparison of the uncertainty scores obtained under the two treat-
ments was also made on a sentence-to-sentence basis. This also proved
non—significant. Both analyses relevant to this hypothesis show that
the introductory paragraph had no significant effect on the uncertainty

scores .

Effects of Semantic Variability (Hypothesis 3)

 

Given the different methods of constructions for Message One and
Message Two, it was hypothesized that the uncertainty scores obtained

under Message Two treatments would be greater than those obtained

29

under Message One treatments. Because the semantic partials used to
construct Message One were related while those for Message Two were
independent, it was felt that the semantic or lexical variability of
Message Two would be greater. It was supposed this increased lexical
variability would increase uncertainty scores.

The uncertainty scores obtained under Message One were combined
with those obtained for Message Two in a joint distribution and ranked.
A Mann—Whitney test was applied to these rankings and found non-signi-
ficant (U = 12, p,> .05). Thus, according to this analysis, the third

hypothesis is not confirmed.

Relationship Between Uncertainty and Learning (Hypothesis 4)

Several analyses were performed on the data in Table l to illum-
inate relations between language uncertainty and learning. Test scores
for the intact messages were first compared with uncertainty scores
over the disorganized message versions. Because test questions were
cast in the same order as sentences of the disorganized messages, it

TABLE 3: KRUSKAL-WALLIS ONE—WAY ANALYSIS OF VARIANCE OF
DISORGANIZED MESSAGES COMPARED WITH LEARNING TESTS

 

 

 

Treatments
Rank Sums D L H DI LI H
Message One 147 63 10.08** 127 83 2.7(N.S.)
Message Two 134 76 4.8* 138 72 6.2*

:'::': P < .01
3" p < .05

30

was felt that if subjects learned nothing by reading the intact mes-
sages, there would be no difference between these sets of scores.

Table 3 summarizes the results of a Kruskall—Wallis two—way anal-
ysis of variance between each disorganized message and the test over
its intact counterpart. For three of the four pairs, this test indi-
cates that the scores are drawn from significantly different popula-
tions. The rank totals (on which this test statistic is based) show
that all differences are in the predicted direction.

The sign test was also applied to these pairs to determine whether
sentence to sentence differences were significant. In all but one
case (TB—T11), this test reveals a significant difference between
uncertainty scores obtained for a disorganized message version and the
test over its intact counterpart. Again all differences are in the

predicted direction.

TABLE 4: KENDALL RANK CORRELATIONS: MESSAGE
ONE TREATMENTS x SENTENCES

T2 T3 T4 T5 T6

T1: Organized .80 .96 .71 .69 .78
T2: Disorganized .69 .51 .69
T3: Organized (Intro) .60 .69
T4: Disorganized (Intro) .42* .64
T5: Learning Test .64
T6: Learning Test (Intro)

 

*N.S. (all other correlations are significant, p <:.05)

31

With respect to learning, it was also argued that the uncertainty
scores obtained for the mutilated messages should predict learning
difficulty for items in the intact text. If this were true, one should
find a significant correlation between uncertainty scores for mutilated
messages and corresponding test items. Table 4 and Table 5 show Ken—
dall rank correlations between sentence uncertainty scores for all
treatments on Message One and Message Two respectively (parametric

correlations are found in Appendix III).

TABLE 5: KENDALL RANK CORRELATIONS: MESSAGE
ONE TREATMENTS x SENTENCES

 

T8 T9 T10 T11 T12

T7: Organized .00 .60* .38 .56* .51*
T8: Disorganized .11 .07 -.20 -.42
T9: Organized (Intro) .42 .20 .20
T10: Disorganized (Intro) .11 .38
T11: Learning Test .51*

T12: Learning Test (Intro)

 

* p <:.05

The fourth hypothesis predicts a significant positive correla—
tion between the uncertainty presented by items of information in a
message and the learning of those items. For Message One, the only
correlation which is non—significant is between the sentences of a dis-

organized message version and test scores over the intact message.

32

Because it is argued in this study that organized message estimates
of uncertainty will be better predictors of message uncertainty, this
finding is in line with the predictions of this study.

For Message Two, the correlations (Table 5) are both more and
less encouraging. These correlations are more encouraging because
the only significant ones are between organized message versions and
test items.

They are less encouraging because the organized message

with an introduction does not correlate with test items.

 

TABLE 6: KENDALL RANK CORRELATION: MESSAGE ONE
TREATMENTS x MESSAGE TWO TREATMENTS
11.18.12.291qu
Tl: Organized 0.00
T2: Disorganized -.033
T3: Organized (Intro) 0.00
T4: Disorganized (Intro) 0.29
T5: Learning Test —.0.07
T6: Learning Test (Intro) 0.29

 

All correlations are non—significant.

Because the deletion patterns
Two are identical, one might suppose
for the various message versions are

schemes. If this were so, one would

the different messages to correlate with one another.

for both Message One and Message
that the uncertainty scores obtained
a function only of the deletion
expect the scores obtained under

Table 6 shows

33

the Kendall rank correlations between Message One and Message Two
counterparts. None of these correlations are significant.

The fourth hypothesis for this study suggests that learning
and uncertainty are closely related. Three separate analyses of data

obtained for this study confirm the hypothesis.

CHAPTER 4

DISCUSSION

Information Theories and Language Processing_

 

 

The research conducted in this study bears upon several impor-
tant theoretical and methodological problems of human communication
and language behavior. Several scientists, notably MacKay (1950,

1951, 1955, 1969) and Bar-Hillel (1964) have constructed models for
human language processing and/or learning based upon Shannon's math-
ematical theory of communication. Generally, these models suggest
that human information processing is a function of organization and
the statistical characteristics of either stimuli or messages. Fur-
ther, these models imply that both degree of organization and the
statistical properties of messages can be tapped by Shannon's lOga-
rithmic measure of information (or uncertainty). The findings of
this study provide evidence of strong relations between information
and uncertainty processes and both message organization and learning.
Moreover, they indicate that the Shannon measure is not only useful
for research into human information processing, but also sensitive to
a variety of message manipulations.

Whereas theories of human information processing have been avail—
able for some time, little research into these processes has been con-.
ducted. This shortcoming has been due largely to a lack of sufficiently
powerful methods for analyzing and manipulating the semantic structure

of messages. The methods developed to construct messages for this

35

study are powerful, both for constructing messages and for analyzing
extant messages. The methods used here for paragraph organization can
be considered a first approximation of a grammar for messages which
goes beyong the level of sentences. This approach should extend con-
siderably the range of message research that can be attempted as well
as the precision with which it can be conducted, for it permits inspec-
tion of aspects of sentences and messages that are not easily manipu-
lated by the methods of traditional and modern sentence grammars.

While these claims may seem somewhat far reaching, a fuller dis—
cussion and interpretation of the results of this study should provide

support for their validity.

Message Organization

 

Previous research has shown a relation between sentence order
and comprehension. Darnell (1963), for instance, constructed six mes-
sages each of which presented an increasing degree of disorder. He
found that the degree of disorganization was positively correlated with
the difficulty of replacing deleted items. The present study is both
a replication and an extention of Darnell's results.

The first hypothesis of the present study predicted that uncer-
tainty scores obtained for disorganized message versions would be signi-
ficantly higher than the scores obtained for organized message versions.
This prediction was strongly confirmed. The total uncertainty scores
obtained for disorganized message versions are significantly higher
than the scores obtained for organized message versions. To this

extent, the present study is only a replication of the Darnell study

36

with two exceptions. First, the organized version of Darnell's mes-
sage was arrived at by constructing a deductive argument. While both
deductive and inductive arguments do appear in texts, they do not
always appear in a straightforward manner. The methods devised for
this study are not limited to any particular argumentative form. Thus,
they are more general. Secondly, the Darnell study employs an n32.
word deletion scheme which does not permit one to tap the exclusively
semantic dimensions of the message.

In going beyong the Darnell study, the present study also shows
that the uncertainty scores obtained for individual sentences are sig—
nificantly lower for organized message versions. Where several studies
have shown that the disruption of normal sentence organization effects

sentence comprehension, this study suggests that the interrelations

 

between sentences might also effect sentence comprehension. Studies

by Marks and Miller (1954), Herriot (1967), Bever, Lackner and Kirk
(1969), Hakes and Foss (1970), and Levelt (1970) show that variations
in the grammar of sentences affect ease of processing or learning. The
present study shows that message processing ease is more than a func—
tion of grammar.

The results of this study suggest that one can organize mes-
sages so as to significantly reduce the amount of presented uncertainty.
By extending the line of research begun here to a greater range of
messages and types of organization, it should be possible to determine
what kinds of message organization will achieve optimum uncertainty
reduction for the recipients of these messages. Research along these
lines must, however, impose greater control on the research design.

Several questions are left unanswered concerning relations between

37

message organization and uncertainty. Some of these questions are:

1. How do uncertainty scores obtained with organized and disorgan—
ized message versions compare with uncertainty scores that could
be obtained if the sentences were presented independently of one
another?

2. How are the uncertainty scores influenced by the prior knowledge
of subjects?

3. How do different individual cognitive styles effect how people
respond to mutilated messages?

4. How do factors of personality and intelligence relate to different
kinds of organizational schemes?

These are only a few of the questions which need to be investigated

by future researchers.

Paragraph Interrelationships

 

Only one previous study dealing with paragraph interrelationships
was found in the literature. Koen, Becker and Young (1969) showed
that subjects were able to recognize paragraph boundaries in unin-
dented prose. The present study attempted to go one step further
and demonstrate the specific nature of this interrelationship. It was
argued that if two paragraphs in sequence were related, the first par-
agraph in the sequence would have the effect of reducing the uncer-
tainty scores for the second paragraph in two ways. First, it was felt
the language introduced in the first paragraph would constrain the
lexical choices made by subjects as they responded to the second par—

agraph. Second, the first paragraph was designed to present the

38

information necessary to accurately fill in the first deleted semantic
partial. The results, however, failed to support this hypothesis.

The effect of the introductory paragraphs was in some cases to raise
uncertainty scores and in others to lower them. No clear trend emerged.

One reason why this treatment may have failed to produce signi-
ficant results is that only one subject made use of the information
in the introductory paragraph to accurately supply the first semantic
partial for the second paragraph. Also, there may possibly have been
some kind of interaction between the writing styles for the two para-
graphs. The introductory paragraphs were written in a more normal style
than were the mutilated paragraphs. Whereas the mutilated passages were
built around the four partials per sentence organizational scheme, the
introductory paragraphs were not. No constraints were imposed on the
writing of introductory paragraphs except the need to include the first
semantic partial.

Although this study yielded no clear results concerning para-
graph interrelationships, some suggestions for future research can be
made. Any future research into paragraph interrelationships should
more accurately specify the nature of the paragraph relationships. The
method used to construct the messages for this study could and should
have been used to analyze the introductory paragraphs to determine whe-
ther and to what extent the semantic partials presented by the introduc—
tory paragraphs were related to those in the mutilated messages. Also,
the test hypothesis should not be directional. If the semantic partials
for the introductory paragraph are largely independent of those in
the second paragraph, the level of uncertainty may be raised for both

messages. Finally, some assessment should be made of the uncertainty

39

presented by the introductory paragraphs before they are coupled with

the mutilated passages.

Semantic Variability

 

Two previous studies (Rosenberg, 1969; Perfetti, 1969) show that
the ability of people to comprehend language is a function of the
meanings available to them. Rosenberg provides evidence that when
sentences are semantically well integrated, subjects will be better
able to recall the accompanying verbal material. Perfetti employs
two measures of semantic variability: (l) the number of semantic
partials, and (2) the number of different words from semantic form
classes. Perfetti presents evidence that both of his semantic mea-
sures better index comprehension that do other measures of sentence
comprehension (e.g., syntactic complexity), and that of these two mea—
sures, the semantic partial is the better index.

The two messages constructed for this study presented the same
number of semantic partials. The messages varied only as to the number
of different lexical choices making up those partials. Message One
partials use 14 different words and Message Two partials use 24. For
this reason, it was predicted that Message Two uncertainty scores would
be greater than Message One scores. The results fail to confirm this
prediction. Still, several possible limitations of the present study
suggest this hypothesis should not be ignored in future research.

One explanation of the failure to support this hypothesis is
that the deletion schemes constructed for the two messages were iden-

tical. Each deletion for a sentence was the deletion of one semantic

4O

partial. The semantic partials for Message Two were designed to be
independent of one another in terms of lexical choices. Message One
partials were designed so that each pair (AB, BC, . . ., LM) was
related by the lexical choices. It was this difference in the con—
struction of the partials which was intended to bring about the
increased variability of the Message Two set. Now, unless the dele—
tion scheme tapped this difference between the messages, one would not
eXpect the uncertainty scores to reflect the differential construction.
For a deletion scheme to tap this dimension of the messages, sequential
pairs of partials must be deleted (e.g.; AB, CD, . . ., LM). The
deletion scheme actually employed taps only one pair, AB.

The fact that the differential aspects of message construction
were not tapped by the deletion scheme gives rise to a second reason
why this hypothesis merits further study. The interdependence of the
semantic partials for Message One operates throughout the message only
in the organized versions. If one looks at the uncertainty score dif—
ferences between the messages for organized versions only, the Mes—
sage One scores are significantly lower than those for Message Two
(Mann—Whitney U = l, p‘< .05). So, when one looks at semantic varia—
bility in conjunction with the organizational pattern which gives rise
to it, semantic variability does seem to have a significant effect.

Future research should also consider the source of semantic
variability. While Perfetti argues that semantic variability is a
function of number of lexical choices, the presented research has
posited that semantic variability is probably a function of the number
of different meanings a subject might have for a word. One can argue

from this perspective that there is no a_priori reason to expect 14

41

different words to have fewer meanings than 24 different words.

Uncertainty and Learning_

 

Prior research by Landahl (1941a, 1941b), Lenneberg (1957), and
Rapoport (1956, 1960, 1961) using artificial languages indicated a
possible relation between the uncertainty of a language and the dif—
ficulties people had learning the language. Because of the artifi-
cial nature of these languages and the many points on which they were
different from natural language, the present study was designed to
determine whether these same processes operate for natural language.
The fourth hypothesis for this study concerned the extent to which
processes of learning are related to uncertainty reduction.

Whereas one can accurately calculate the presented uncertainty
of an artificial language (the number of referents for each symbol is
known), the same is not true for natural languages. Some means for
estimating the uncertainty of a message has to be constructed. The
mutilated, disorganized messages for this study were used to index how
much uncertainty was presented by the individual sentences. The organ-
ized message versions were used to index the extent to which an organ—
ized pattern for these sentences would reduce that uncertainty. Keep-
ing these points in mind, it is possible to consider the hypothesis that
learning is a function of uncertainty reduction.

To test whether learning is a process of uncertainty reduction,
subjects were first asked to read the intact message versions and then
respond to a test covering those messages. The tests for this study

were designed to ask questions calling for subjects to identify semantic

42

partials which had been deleted from the mutilated messages. The
questions were presented in the same order as the sentences of the muti-
lated message versions. It was argued that if the uncertainty scores
obtained for responses to test questions were lower than the scores
obtained for disorganized message versions, these lower scores must

have resulted from what subjects learned. The data confirmed this
prediction, both with respect to total message scores and on a sentence-
to-sentence basis. In fact, inspection of uncertainty scores obtained
for test questions indicated they were in all cases lower than the
scores obtained from organized message versions.

A more powerful aspect of the relationships between learning and
uncertainty is revealed by the intercorrelations between uncertainty
scores for sentences under the different treatments. It was hypothesized
that the correlations between uncertainty scores for sentences of organ-
ized message versions and test items would be better than those between
disorganized versions and test items. For Message One, both organized
and disorganized message versions were found to correlate significantly
with test items. For Message Two, only organized versions (though not
all) correlated significantly with test items. Even though some dis—
organized message versions did correlate with test items, it can be
shown that organized message versions were significantly better pre—
dictors of learning difficulty. Tables 4 and 5, found in Chapter 3, show
eight organized/disorganized pairs of correlations. Of these eight
pairs, only one shows a higher correlation with a test for a disorgan—
ized as compared with its organized counterpart. Under the sign test,
the probability of such an event is less than .05. In other words, the

uncertainty scores obtained from organized message versions are signifi—

43

cantly better predictors of learning difficulty than those obtained from
disorganized messages.

A third aspect of relations between uncertainty and learning
has to do with the source of uncertainty. One possible interpretation
for the findings of this study is that the obtained uncertainty scores
are a function of how many times a semantic partial appears in the text,
whether mutilated or intact, before its deletion. In other words, the
uncertainty score for a semantic partial may be related only to its
frequency of appearance in the test. If this were the case, one would
expect the scores for sentences in the two messages to be significantly
correlated. Because both the organizational schemes and deletion
schemes are identical, this is a reasonable exPectation. None of the
correlations between a Message One and a Message Two counterpart are
significant. In fact, five of these six correlations (Table 6) are
near zero or negative. This finding would seem to support MacKay's
(1969) argument that message uncertainty is a function of the meanings
triggered in people by messages.

While it is argued throughout this study that uncertainty is a
function of semantic variability, the results by no means conclusively
support this argument. Though the results are generally supportive,
further research is needed.

The present study, where it considers relations between uncer-
tainty and learning, has several weaknesses. First, a better index of
the uncertainty of the messages could be obtained by having different
groups of people respond to the mutilated texts in such a way that an
index of uncertainty for each partial could be obtained. Such a pro-

cedure would permit one to get a clearer view of how the partials are

44

related one to another. Second, uncertainty indices should be obtained
for each sentence as it stands alone. As this study was conducted,

one could argue there is no basis for saying that uncertainty was
reduced by organizing the sentences. The converse could be true. The
disorganized message versions may have produced greater uncertainty
because of the disorganization such that scores for sentences of dis—
organized versions might be higher than scores obtained for sentences
standing alone. These possibilities should be investigated in future

research.

Summary

This study explores four general questions: (1) how is message
organization related to uncertainty? (2) how do paragraphs in sequence
effect uncertainty? (3) how is semantic variability related to uncer—
tainty? and (4) how is message uncertainty related to processes of
learning? The evidence provided by this study provides a partial answer
to the first question. A decrease in message organization clearly
increases the uncertainty presented by messages. With respect to the
second question, no clarification is provided by this study. One can
only speculate that paragraph interrelationships are not simple. What—
ever the interrelationships, they are more complex than anticipated.

The third question is also not clearly answered. While the study fails
to support the stated hypothesis, the failure is to some extent a failure
of the design. The fourth question, concerning relations between learn—
ing and uncertainty, is confirmed for this study. Learning does seem

to be a process of uncertainty reduction. Also, the uncertainty of

45

messages and/or sentences does provide an index of learning difficulty.
Whereas the findings of this study are encouraging, further
research is indicated and necessary. Although procedures are designed
to obtain information values for natural language messages, the
design of the messages employed in this research is somewhat artifi—
cial. Rarely, if ever, do our everyday messages take the overbear-
ingly redundant form of the messages used here. There is also the
question of just how meaningful the messages employed here are. Other,
more specific, weaknesses have already been pointed out.
The above comments are by no means meant to denigrate the find-
ings of this research. It only seems necessary to be careful how the

results are interpreted until more research has been conducted.

BIBLIOGRAPHY

 

 

.l. Ill—I. I III III I ’ltlllln I I

BIBLIOGRAPHY

I Ammons, R. B., "Effects Of Knowledge On Performance: A Survey And
‘ Tentative Formulation,” JOURNAL OF GENERAL PSYCHOLOGY. Vol. 54,
1956, pp. 279-299.

ifAttneave, P., APPLICATIONS OF INFORMATION THEORY TO PSYCHOLOGY. Holt,
“ Rinehart and Winston, New York, 1967.

Bar-Hillel, Y., LANGUAGE AND INFORMATION. Addison-Wesley, London, 1964.

Bever, T. C., Lackner, J. R. and Kirk, R., "The Underlying Structure Of
Sentences Are The Primary Units Of Immediate Speech Processing," PER-
CEPTION AND PSYCHOPHYSICS. V01. 5, 1969, pp. 225-234.

Born, M. A., Rubenstein, H. and Sterling, T. C., "Sources Of Contextual
Constraint Upon Words In Sentences," JOURNAL OF EXPERIMENTAL PSYCHO-
PHYSICS. Vol. 57, 1959, pp. 171-180.

Brillouin, L., "Mathematics, Physics, And Information," INFORMATION AND
CONTROL. v61. 1, 1957, pp. 1-5.

Butler, B. E. and Merikle, P. M., ”Uncertainty And Meaningfulness In A
Paired-Associate Learning," JOURNAL OF VERBAL LEARNING AND VERBAL
BEHAVIOR. Vol. 9, 1970, pp. 634-641.

Chomsky, N., SYNTACTIC STRUCTURES. Mouton and Co., The Hague, 1957.

Chomsky, N., ASPECTS OF A THEORY OF SYNTAX. M.I.T. Press, Cambridge,
Mass., 1965.

Darnell, D. K., ”The Relation Between Sentence Order And Comprehension,"
SPEECH MONOGRAPHS. Vol. 30, 1963, pp. 97-100.

Fillenbaum, 8., Jones, L. V. and Rapoport, Amnon, "The Predictability Of
Words And Their Grammatical Classes As A Function Of Rate Of Deletion
From A Speech Transcript," JOURNAL OF VERBAL BEHAVIOR. Vol. 2, 1963,
pp. 186-194.

Gross, H. S., "The Effect On Word Recognition Of The Frequency Of Word
Association," LANGUAGE AND SPEECH. Vol. 9, 1966, pp. 52-62.

Hakes, D. T. and Foss, D. J., "Decision Processes During Sentence Com-
prehension: Effects Of Surface Structure Reconsidered," PERCEPTION AND
PSYCHOPHYSICS. Vol. 8, 1970, pp. 413-416.

Herriot, P., "Phrase Units And The Recall Of Grammatically Structured
Nonsence," BRITISH HOURNAL OF PSYCHOLOGY. Vol. 58, 1967, pp. 237—242.

46

Hyman, L. M. and Kaufman, H., "Information And The Memory Span,” PERCEP-
TION AND PSYCHOPHYSICS. Vol. 1, 1966, pp. 235-237.

Kochen, M. and Galanter, E., ”The Acquisition And Utilization Of Informa-
tion In Problem Solving And Thinking," INFORMATION AND CONTROL. Vol. 1,
1958, pp. 267-2880

Koen, F., Becker, A. and Young, R., "The Psychological Reality Of The
Paragraph," JOURNAL OF VERBAL LEARNING AND VERBAL BEHAVIOR. Vol. 8,
1969, pp. 49-53.

Landahl, H. D., ”Studies In The Mathematical Biophysics Of Discrimination
And Conditioning I," BULLETIN OF MATHEMATICAL BIOPHYSICS. Vol. 3, 1941,
pp. 13-26.

Landahl, H. D., "Studies In The Mathematical Biophysics Of Discrimination
And Conditioning II: Special Case: Errors, Trials, And Number of Pos-
sible Responses," BULLETIN OF MATHEMATICAL BIOPHYSICS. Vol. 3, 1941,

pp. 71-77.

Lenneberg, E. H., ”A Probabilistic Approach To Language Learning,” BEHA-
VIORAL SCIENCE. Vol. 2, 1957, pp. l—l2.

Levelt, W. J. M., "Hierarchical Chunking In Sentence Processing," PER-
CEPTION AND PSYCHOPHYSICS. v61. 8, 1970, pp. 99-103.

Luria, A. R., ”Speech Development And The Formation Of Mental Processes,"
A HANDBOOK OF CONTEMPORARY SOVIET PSYCHOLOGY. (M. Cole and I. Maltz—
man, eds.), Basic Books, Inc., New York, 1969.

Luria, A. R., HIGHER CORTICAL FUNCTIONS IN MAN. Basic Books, Inc., New
York, 1970.

MacKay, D. M., ”Quantal Aspects Of Scientific Information," PHILOSOPHY
MAGAZINE. Vol. 41, 1950, pp. 289-311.

MacKay, D. M., "Mindlike Behavior In Artifacts," BRITISH JOURNAL OF THE
PHILOSOPHY OF SCIENCE. Vol. 2, 1951, pp. 105-121.

MacKay, D. M., "Complementary Measures Of Scientific Information-Con-
tent," METHODOS. V61. 7, 1955, pp. 63-90.

MacKay, D. M., INFORMATION, MECHANISM AND MEANING. M.I.T. Press, Cam-
bridge, Mass., 1969.

Malic, D., ”Entropy And Philosophy," PROGRESS OF CYBERNETICS. Vol. 1,
(j. Rose, ed.), Gordon and Breach, London, 1970, pp. 149-154.

Marks, L. E. and Miller, G. A., "The Role Of Semantic And Syntactic Con-

straints In The Memorization Of English Sentences," JOURNAL OF VERBAL
LEARNING AND VERBAL BEHAVIOR. Vol. 3, 1964, pp. 1-5.

47

Mickelson, J., AN APPLICATION OF INFORMATION THEORY TO HUMAN COMMUNI-
CATION BEHAVIOR. Unpublished Masters Thesis, Wichita State Uni., 1958.

Miller, G. A. and Friedman, E. A., "The Reconstruction Of Mutilated
English Texts," INFORMATION AND CONTROL. Vol. 1, 1957, pp. 38-55.

Miller, G. A., Newman, E. B. and Friedman, E. A., "Length Frequency
Statistics For Written English,” INFORMATION AND CONTROL. Vol. 1,
1958, pp. 370—389.

I)“ Moss, D. E. and Neidt, C. 0., ”Applicability Of Information Theory To
“ Learning," PSYCHOLOGICAL REPORTS. V01. 24, 1969, pp. 471-478.

Perfetti, C. A., "Lexical Density And Phrase Structure Depth As Varia-
bles In Sentence Retention," JOURNAL OF VERBAL LEARNING AND VERBAL
BEHAVIOR. Vol. 8, 1969, pp. 719-724.

Pollack, I., "Message Uncertainty And Message Reception III. Effect
Of Restriction Of Verbal Context," JOURNAL OF VERBAL LEARNING AND
VERBAL BEHAVIOR. Vol. 1, 1963, pp. 392-395.

Postman, Leo, "Effects Of Word Frequency On Acquisition And Retention
Under Conditions Of Free Recall Learning," QUARTERLY JOURNAL OF EXPER—
IMENTAL PSYCHOLOGY. Vol. 22, 1970, pp. 185-195.

Rapoport, A., ”What Is Information," ETC. Vol. 10, 1953, pp. 247—260.

Rapoport, A., "On The Application Of The Information Concept To Learn-
ing Theory,” BULLETIN OF MATHEMATICAL BIOPHYSICS. Vol. 18, 1956,
pp. 317-3210

Rapoport, A., "A Derivation Of A Rote Learning Curve From The Total
Uncertainty Of A Task,” BULLETIN OF MATHEMATICAL BIOPHYSICS. Vol. 22,
1960, pp. 85-97.

Rapoport, A., "The Perfect Learner," BULLETIN OF MATHEMATICAL BIO—
PHYSICS. Vol. 23, 1961, pp. 321-335.

Rosenberg, S., "The Recall Of Verbal Material Accompanying Semantically
Well Integrated And Semantically Poorly Integrated Sentences,” JOURNAL
OF VERBAL LEARNING AND VERBAL BEHAVIOR. V01. 8, 1969, pp. 732—736.

Rubenstein, H. and Pollack, I., "Word Predictability And Intelligibil-
ity," JOURNAL OF VERBAL LEARNING AND VERBAL BEHAVIOR. Vol. 2, 1963,
pp. 147-158.

Shannon, C. E., "A Mathematical Theory Of Communication," BELL SYSTEM
TECHNICAL JOURNAL. v61. 27, 1948, pp. 379-423, pp. 623-656.

Shannon, C. E. and Weaver, N., THE MATHEMATICAL THEORY OF COMMUNICATION.
University of Illinois Press, Urbana, Ill., 1957.

Staniland, A. C., PATTERNS OF REDUNDANCY: A PSYCHOLOGICAL STUDY. Cam-
bridge University Press, Cambridge, Ill., 1957.

48

Swanson, J. M. and Wickens, D. D., "Preprocessing On The Basis Of Fre-
quency Of Occurrence," QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY.
Vol. 19, 1969, pp. 378-383.

Tannenbaum, P. H., Williams, F. and Clark, R. H., ”Effects Of Gramma-
tical Information On Word Predictability," JOURNAL OF COMMUNICATION.
Vol. 19, 1969, pp. 41-48.

Taylor, W., "A New Tool For Measuring Readability,” JOURNALISM QUAR-
TERLY. Vol. 30, 1953, pp. 415-433.

Treisman, A. M., "Verbal Responses And Contextual Constraints In Lan-
guage,” JOURNAL OF VERBAL LEARNING AND VERBAL BEHAVIOR. Vol. 4,
1965, pp. 118-128.

Weinstein, R. S., Feldstein, S., and Jaffe, J., "The Correlation Of

Vocal Context And Lexical Predictability," LANGUAGE AND SPEECH.
Vol. 8, 1965, pp. 56-66.

49

APPENDICES

APPENDIX I

SEMANTIC PARTIALS

APPENDIX I

SEMANTIC PARTIALS

Message One

 

 

A. TASK R CONCEPT

B. CONCEPT R UNCERTAINTY

C. UNCERTAINTY R REDUCED

D. REDUCED R INFORMATION

E. INFORMATION R GAINED

F. GAIN R PROCESS

G. PROCESS R LEARNING

H. LEARNING R PATTERN

I. PATTERN R ORGANIZATION OF STIMULI

J. ORGANIZATION R CODE SEQUENCES

K. CODE SEQUENCES R SENTENCES

L. SENTENCES R INFORMATION

M. INFORMATION R UNCERTAINTY
Messaga_Two

A. COMMUNICATION R STUDY

B. PEOPLE R SYMBOLS

C. CODES R ORGANIZED

D. SEQUENCES R SENTENCES

E. INFORMATION R TRANSACTIONS

F. MESSAGES R UNCERTAINTY

G. SEMANTIC R AMBIGUITY

H. MESSAGE RECEIVER R RESOLVE

I. UNDERSTANDING R OCCURS

J. MEANING R ACQUIRED

K. MESSAGE CONTENTS R LEARNED

L. UNCERTAINTY R REDUCTION

M. COMMUNICATION R ACHIEVED

50

APPENDIX II

INSTRUMENTS

APPENDIX II

 

 

 

INSTRUMENTS
Message One: Organized
This involves the uncertainty and
its reduction by information. The states

 

that uncertainty will be reduced as information is gained. We assume

that uncertainty is reduced as is according

 

 

to some mental process. Generally, the takes place as

 

is gained in the process of learning. Information is

 

 

 

 

in the learning as one learns patterns.
That which is gained in the process of learning is the
of . In the of ,

 

 

 

 

patterns organize stimuli into code sequences. The learning of

patterns involves the organization of stimuli into

 

like . This patterning or of

 

 

 

into like sentences facilitates

 

 

transmission of information. Thus, the organization of stimuli into

code sequences facilitate transmission and

 

 

reduction.

51

APPENDIX II

 

 

 

 

 

INSTRUMENTS
Message One: Disopganized
That which is gained in the process of learning is
the of . Generally, the takes
place as is gained in the process of learning. The learn-

 

ing of patterns involves the organization of stimuli into

 

like . The states

 

 

 

that uncertainty will be reduced as information is gained. Thus, the

organization of stimuli into code sequences facilitates

 

 

 

 

transmission and reduction. This involves
the uncertainty and its reduction by information. This
patterning or of into

 

 

 

like sentences facilitates information transmission. We assume that

 

 

 

 

 

uncertainty is reduced as is according to
some mental process. In the of , patterns
organize stimuli into code sequences. Information is in
the learning as one learns patterns.

 

52

 

 

INININI

APPENDIX II

INSTRUMENTS

Message One: Organized With Introductory Paragraph

The study of human communication draws upon concepts from a
variety of different disciplines. Among these are concepts like organ-
ization, memory, information, learning and so on. The task in which
you are here participating is concerned with generating data to help

assess the relationships between several of these concepts and uncertainty.

This involves the uncertainty and its

 

 

reduction by information. The states that

 

uncertainty will be reduced as information is gained. We assume that

 

 

 

 

 

 

 

 

 

uncertainty is reduced as is_ according to some
mental process. Generally, the takes place as

is gained in the process of learning. Information is in
the learning as one learns patterns. That which is gained
in the process of learning is the of

In the of , patterns organize stimuli into

 

 

code sequences. The learning of patterns involves the organization of

stimuli into like . This

 

 

patterning or of into

 

 

 

like sentences facilitates transmission of information. Thus, the

organization of stimuli into code sequences facilitate

 

transmission and reduction.

 

53

APPENDIX II

INSTRUMENTS

Message One: Disopganized With Introductory Paragraph

The study of human communication draws upon concepts from a
variety of different disciplines. Among these are concepts like
organization, memory, information, learning and so on. The task in
which you are here participating is concerned with generating data
to help assess the relationships between several of these concepts

and uncertainty.

 

 

 

 

That which is gained in the process of learning is
the of . Generally, the takes
place as is gained in the process of learning. The learn-

 

ing of patterns involves the organization of stimuli into

 

like . The states

 

 

 

that uncertainty will be reduced as information is gained. Thus, the

organization of stimuli into code sequences facilitates

 

 

 

 

transmission and reduction. This involves
the uncertainty and its reduction by information. This
patterning or of into

 

 

 

like sentences facilitates information transmission. We assume that

 

 

 

 

 

uncertainty is reduced as is according to
some mental process. In the of, , patterns
organize stimuli into code sequences. Information is in
the learning as one learns patterns.

 

54

APPENDIX II

INSTRUMENTS

Messaga One: Intact

 

This task involves the concept uncertainty and its reduction
by information. The uncertainty concept states that uncertainty will
be reduced as information is gained. We assume that uncertainty will
be reduced as information is gained according to some mental process.
Generally, the reduction takes place as information is gained in the
process of learning. Information is gained in the learning process
as one learns patterns. That which is gained in the process of learn-
ing pattern is the organization of stimuli. In the process of learning,
patterns organize stimuli into code sequences. The learning of patterns
involves the organization of stimuli into code sequences like sentences.
This patterning or organizing of stimuli into code sequences like
sentences facilitates transmission of information. Thus, the organi—
zation of stimuli into code sequences facilitate information trans-

mission and uncertainty reduction.

55

APPENDIX II

INSTRUMENTS

Message One: Intact With Introductory Paragraph

 

The study of human communication draws upon concepts from a
variety of different disciplines. Among these are concepts like
organization, memory, information, learning and so on. The task in
which you are here participating is concerned with generating data
to help assess the relationships between several of these concepts

and uncertainty.

This task involves the concept uncertainty and its reduction by
information. The uncertainty concept states that uncertainty will be
reduced as information is gained. We assume that uncertainty will
be reduced as information is gained according to some mental process.
Generally, the reduction takes place as information is gained in the
process of learning. Information is gained in the learning process
as one learns patterns. That which is gained in the process of learn-
ing pattern is the organization of stimuli. In the process of learn-
ing, patterns organize stimuli into code sequences. The learning of
patterns involves the organization of stimuli into code sequences
like sentences. This patterning or organizing of stimuli into code
sequences like sentences facilitates transmission of information.
Thus, the organization of stimuli into code sequences facilitate

information transmission and uncertainty reduction.

56

APPENDIX II

INSTRUMENTS

MessagaaOne: Test

 

The following questions concern the paragraph you have just read:

 

 

 

 

 

1. As we learn , we gain the of
2. When is gained, uncertainty is .
3. Stimuli are organized into much like

 

 

 

4. The statement ”uncertainty will be reduced as information is

gained" refers to the .

 

5. transmission and reduction are facili—

 

 

tated by the organization of stimuli into code sequences.

6. The paragraph describes a involving a certain

 

 

7. Information transmission is facilitated by

 

 

into .

 

8. There is some mental process involved when uncertainty is reduced

and is .

 

 

9. Patterns organize stimuli into code sequences in the

 

of .

 

10. In the learning , information is as one

 

 

learns pattern.

57

APPENDIX II

 

 

 

INSTRUMENTS
Message Two: Organized
involves the of how people use symbols
or codes organized into sequences like sentences. use

 

, codes and organized sequences or sentences in informa-

 

tion transactions. When codes are organized in sequences like sen-

tences for , the resulting messages produce

 

uncertainty. The or used in information

 

 

transactions produce uncertainty or semantic ambiguity. Thus, infor—

mation transactions involve which produce

 

 

and create semantic ambiguity which message receivers must resolve.
Message uncertainty or semantic ambiguity must be resolved by the

message receiver before can . As

 

 

 

is resolved by the message receiver, understanding occurs

 

and meaning is acquired. Through message receiver resolution, under-

standing occurs, meaning is acquired and

 

 

are . With the occurrence of understanding and

 

 

by learning of message contents, uncertainty is reduced.

 

As meaning is acquired through learning of message contents and reduction

of uncertainty, is .

 

 

58

APPENDIX II

INSTRUMENTS

Message Two: Disorganized

 

Message uncertainty or semantic ambiguity must be resolved by

the message receiver before can . The

 

 

or used in information transactions produce

 

 

uncertainty or semantic ambiguity. Through message receiver resolu-

tion, understanding occurs, meaning is acquired and

 

are . use , codes

of

 

 

 

or organized sequences, or sentences in information transactions. As.
meaning is acquired through learning of message contents and reduction

of uncertainty, is . involves

 

 

 

the of how people use symbols or codes organized into.

 

sequences like sentences. With the occurrence of understanding and

by learning of message contents, uncertainty

 

is reduced. When codes are organized in sequences like sentences for

, the resulting messages produce uncertainty.

 

As is resolved by the message receiver,

 

understanding occurs, meaning is acquired and message contents are

learned. Thus, information transactions involve which

 

produce and create semantic ambiguity which message

 

receivers must resolve.

59

 

APPENDIX II

INSTRUMENTS

Message Two: Organized With IntroductoryaParagpaph

The task in which you are here participating is concerned with
generating data to help assess the relationships between several com-
munication concepts and uncertainty. Among these are concepts like
"code," information, learning and so on. The study of communication

draws upon many of these concepts.

 

 

involves the of how people use symbols

or codes organized into sequences like sentences. use

 

, codes and organized sequences or sentences in informa-

 

tion transactions. When codes are organized in sequences like sen-

tences for , the resulting messages produce

 

 

uncertainty. The or used in information

 

 

transactions produce uncertainty or semantic ambiguity. Thus, informa—

tion transactions involve which produce and

 

 

create semantic ambiguity which message receivers must resolve. Message
uncertainty or semantic ambiguity must be resolved by the message

receiver before can. . As

 

 

 

is resolved by the message receiver, understanding occurs

 

and meaning is acquired. Through message receiver resolution, under-

standing occurs, meaning is acquired and

 

are . With the occurrence of understanding and

 

 

by learning of message contents, uncertainty is reduced.

 

As meaning is acquired through learning of message contents and reduction

of uncertainty, is .

 

 

60

APPENDIX II

INSTRUMENTS
Messaga Two: Disorganized With Introductory_Paragpaph

The task in which you are here participating is concerned with
generating data to help assess the relationships between several com-
munication concepts and uncertainty. Among these are concepts like
”code,” information, learning and so on. The study of communication

draws upon many of these concepts.

Message uncertainty or semantic ambiguity must be resolved by

the message receiver before can . The

 

 

or used in information transactions produce_

 

 

uncertainty or semantic ambiguity. Through message receiver resolution,

understanding occurs, meaning is acquired and

 

are . use , codes or organized

a

 

 

 

sequences, or sentences in information transactions. As meaning is

acquired through learning of message contents and reduction of uncer-

tainty, is . involves the

 

 

 

of how people use symbols or codes organized into sequences

 

like sentences. With the occurrence of understanding and

 

by learning of message contents, uncertainty is reduced.

 

When codes are organized in sequences like sentences for

 

, the resulting messages produce uncertainty. As

 

 

is resolved by the message receiver, understanding occurs,

 

meaning is acquired and message contents are learned. Thus, informa-

tion transactions involve which produce and

 

 

create semantic ambiguity which message receivers must resolve.

APPENDIX II

INSTRUMENTS
Message Two: Intact With Introductory Paragraph

The task in which you are here participating is.concerned with
generating data to help assess the relationships between several com—
munication concepts and uncertainty. Among these are concepts like
"code," information, learning and so on. The study of communication

draws upon many of these concepts.

Communication involves the study of how people use symbols or
codes organized into sequences like sentences. People use symbols,
codes and organized sequences or sentences in information transactions.
When codes are organized in sequences like sentences for information
transactions, the resulting messages produce uncertainty. The sequences
or sentences used in information transactions produce uncertainty or
semantic ambiguity. Thus, information transactions involve messages
which produce uncertainty and create semantic ambiguity which message
receivers must resolve. Message uncertainty or semantic ambiguity
must be resolved by the message receiver before understanding can occur.
As semantic ambiguity is resolved by the message receiver, understand-
ing occurs and meaning is acquired. Through message receiver resolu-
tion, understanding occurs, meaning is acquired and message contents
are learned. With the occurrence of understanding and meaning acquisi-
tion by learning of message contents, uncertainty is reduced. As
meaning is acquired through learning of message contents and reduction

of uncertainty, communication is achieved.

63

APPENDIX II

 

 

 

INSTRUMENTS
Message Two: Test
1. can only after semantic ambiguity is
resolved.
2. Uncertainty is produced by the or used

10.

 

 

in information transactions.

As are meaning is acquired,

 

 

understanding occurs through message receiver resolution.

 

 

 

 

Codes, organized sequences or sentences and are used
by in information transactions.
is as meaning is acquired through learn-

ing or the reduction of uncertainty.
The uses people make of symbols or codes organized into sequences

like sentences are involved in the of .

 

 

Uncertainty is reduced along with the occurrence of understanding

and by learning of message contents.

 

Messages produce uncertainty when codes are organized in sequences

like sentences for .

 

 

Meaning is acquired and message contents learned as

 

is resolved by the message receiver.

 

which produce and create semantic ambi—

 

 

guity which must be resolved by message receivers are involved in

information transactions.

64

IIIIIII‘I

APPENDIX III

PARAMETRIC ANALYSES

APPENDIX III

PARAMETRIC ANALYSES

ANALYSIS OF VARIANCE (MODEL II):
TREATMENTS BY MESSAGES

 

 

Source 83 df MS F
Columns (Treatments) 76.7632 5 15.3526 (5.9 )*
Rows (Messages) 5.6606 1 5.6606 (2.04)
Interaction 12.7612 5 2.5523
Error 347.7001 168 2.0696
TOTALS 442.8855 179

 

* (p <:1001) Error and interaction terms are pooled to arrive at F.

PEARSON PRODUCT MOMENT CORRELATION:
MESSAGE ONE TREATMENTS X SENTENCES

 

Tl: Organized .81 .85 .76 .84 .90
T2: Disorganized .89 .94 .66 .73
T3: Organized (Intro) .90 .72 .75
T4: Disorganized (Intro) .60 .74-
T5: Learning Test .86

T6: Learning Test (Intro)

65

APPENDIX III

PARAMETRIC ANALYSES

PEARSON PRODUCT MOMENT CORRELATION:
MESSAGE TWO TREATMENTS x SENTENCES

T8 T9 T10 T11 T12

T7: Organized -.03 .79 .61 .61 .60
T8: Disorganized .32 .15 -.34 -.52
T9: Organized (Intro) .83 .22 .34
T10: Disorganized (Intro) .24 .64
T11: Learning Test .62

T12: Learning Test (Intro)

PEARSON PRODUCE MOMENT CORRELATION: MESSAGE
ONE TREATMENTS x MESSAGE TWO TREATMENTS

T7 T8 T9 T10 T11 T12

Tl: Organized -.04

T2: Disorganized -.39

T3: Organized (Intro) .01

T4: Disorganized (Intro) .44

T5: Learning Test .07

T6: Learning Test (Intro) .32

66

MICHIGAN STQTE U

III“
:31

IB

NIV. L RnRIES
AllIIIIIIIIIIIIIIIIIIIIIII
0 66

38343

IIII

2931